Graduate School of Informatics

This research evaluates the convergence of High-Performance Computing (HPC), Big Data, Artificial Intelligence (AI), and Cloud Computing technologies using bibliometric analysis, including performance and network analysis. The results reveal a rapidly growing literature with a significant increase in research activities in recent years, identifying key trends and patterns in the literature, including top published authors, most productive institutions, cited articles, and influential publications. This thesis provides valuable insights by identifying the bibliometric trends across the concept of technological convergence of HPC-Big Data-AI-Cloud Computing technologies, which is important for both academia and industry.

Date: 24.01.2023 / 11:00 Place: A-212

Catastrophic forgetting is common in the connectionist models while learning from a sequence of tasks. This study aims to investigate different continual learning methods on face analysis tasks involving age estimation, gender recognition, emotion recognition, and face recognition.  We analyze face analysis in two stages, which is also very common in Artificial Neural Networks: face detection and face attributes analysis. Firstly, experiments for learning face detection and facial landmark detection are conducted by studying multitask learning. Secondly, some continual learning methods inspired by biological systems are leveraged to overcome catastrophic interference in artificial models.

Date: 26.01.2023 / 09:30 Place: B-116

The current thesis compares the communication breakdown repair strategies of adults with High Functioning Autism Spectrum Disorder (HFASD) and age- and education-matched healthy individuals. To create a discourse corpus, video recordings of ASD and control group members performing a joint task under two conditions—matched with the acquainted or the experimenter—were analyzed. This experimental setup creates a situation where participants instruct and follow each other to examine speech repair strategies and elaboration needs. The study's transcripts were content-analyzed, and participants' repair utterances were marked and statistically analyzed.

Date: 26.01.2023 / 14:00 Place: B-116

Bouldering is a subdiscipline of climbing that challenges both problem-solving skills and physical abilities. This study focuses on the decision-making processes of climbers when solving boulder problems in an artificial and standardized climbing wall called MoonBoard. This study aims to build a goal-based AI agent that learns from previous solutions to plan the sequence of actions for novel boulder problems it encounters. We evaluate the agent's cost estimates and climbing solutions by comparing it to the difficulty estimations and solutions provided by expert climbers.

Date: 26.01.2023 / 09:30 Place: A-212

This study aims to gain an understanding of the content and the epistemic causal connective acquisition process of children aged 6;5- 8 in Turkish. For this purpose, to test whether there are connectives that children use specific to domains of causality as adults do (Çokal, Zeyrek, & Sanders, 2020); child and adult participants are given both descriptive (biased for content relations) and argumentative (biased for epistemic relations) tasks. Data that is collected from these tasks are annotated and statistically analyzed.

Date: 26.01.2023 / 13:30 Place: A-212

This research proposes a new AI-based approach to improve the user experience on online marketplaces by validating product images and metadata in an automated fashion. As e-commerce has become popular, online marketplaces have seen a surge in merchants offering various products and maintaining data quality of product metadata and images can be challenging. Our approach offers several advantages over traditional methods, including handling complex and noisy data and adapting to various challenging product categories, such as fashion items. The effectiveness of this approach is demonstrated through comparisons with traditional methods and in different settings.

Date: 23.01.2023 / 13:00 Place: A-212

Counterfactual statements are conditional statements with false or unrealized antecedents. This thesis aims to analyze counterfactuals in Turkish using linguistic and causality perspectives, which will shed new light on their nature. The research will examine the antecedent’s need to be false or unrealized, the use of a complex suffix in Turkish and its relation to counterfactual interpretation, the difference between -DIysA and -sAydI and the role of pragmatics in the interpretation of a counterfactual statement. Additionally, this thesis is the first to have a clear focus on Turkish counterfactuals and use directed acyclic graphs to graphically represent counterfactual scenarios.

Date: 26.01.2023 / 11:00 Place: A-212

The goal of this thesis is to evaluate the skills mismatch in the curricula of HPC MSc's programs for some of the most relevant positions in the HPC areas in light of the ESCO database criteria and industry requirements. The four separate profiles "Data Science, Computer Architecture, Parallel Programming, and DevOps" that are thought to be crucial in HPC are the focus of this research. The methodology of this contribution explicitly examines the key responsibilities for the aforementioned positions and conducts a gap analysis based on Natural Language Processing (NLP) techniques for the competencies required for each in the MSc degree curriculum. The goal of applying NLP is to determine the degree to which the occupational skills stated in the ESCO and the HPC graduate courses’ syllabuses are semantically similar, as well as the level at which they overlap.

Date: 16.01.2023 / 13:00 Place: B-116

This thesis demonstrates research towards innovating a novel framework that (i) could be trained by supervised learning of discourse parsing from annotated and parsed relations, (ii) makes it easy to build end-to-end shallow discourse parsing system, (iii) forms a discourse relation detector benchmark for low resourced languages. Our modern neural network approach, integrated with contextualized text embedding, produced by pretrained language models, accomplished two sub-tasks of shallow discourse parsing, namely, identification of discourse relation realization types and the sense classification of explicit and implicit relations. The effect of multilingual data aggregation on the classification of discourse relation type through Cross-lingual Transfer Learning experiments is researched too.

Date: 25.01.2023 / 10:00 Place: A-108